I wonder if the Faiss team is planning something similar. In particular, I wonder if the following technical context can make the PyPI story more accessible for Faiss and its users than before.
(1) NVIDIA now publishes its packages on PyPI, and my understanding is that PyTorch already relies on that.
(2) uv is becoming increasingly popular in the Python community, and it provides an easy way to specify custom PyPI index for specific packages in pyproject.toml, if, for whatever reason, the default index is not suitable for a given package. For example:
The PyTorch team will soon stop publishing packages to Conda: https://github.com/pytorch/pytorch/issues/138506
I wonder if the Faiss team is planning something similar. In particular, I wonder if the following technical context can make the PyPI story more accessible for Faiss and its users than before.
(1) NVIDIA now publishes its packages on PyPI, and my understanding is that PyTorch already relies on that.
(2) uv is becoming increasingly popular in the Python community, and it provides an easy way to specify custom PyPI index for specific packages in
pyproject.toml
, if, for whatever reason, the default index is not suitable for a given package. For example: